Public Benchmarks Enable Private Evaluations

The core asset here is not the software dashboard, it is trusted ground truth. Enterprise buyers will only pay for a private eval if they believe the scoring system reflects real work, and public benchmarks are how that trust gets built in the open. Vals turns benchmark publishing into proof of methodology, then sells private runs on customer data, domain specific rubrics, and human review workflows that feel less like generic analytics and more like outsourced testing infrastructure.

• Vals publishes domain benchmarks in finance, law, tax, mortgage, and software, with model rankings, update dates, and task level methodology. That public track record shows buyers how tasks are structured before they hand over proprietary workflows for a private evaluation project.
  1 vals 2 github 3 vals
• This is different from LangSmith and Braintrust, which mainly help teams run internal eval loops on their own apps. Those tools are strong for regression testing and observability, but they do not create the same public reference layer that makes a third party scorer look authoritative across companies.
  4 sacra 5 braintrust
• Comparable companies are converging on the same pattern. Arena uses public leaderboards to expand into paid vertical evaluations, and Surge AI has built public benchmarks to demonstrate its evaluation quality. That suggests benchmark publication is becoming the go to trust layer for selling higher value enterprise testing work.
  6 langchain 7 sacra 8 sacra

The next step is a split market, where public benchmarks become the storefront and private eval systems become the revenue engine. Companies that can keep publishing respected domain tests, while also turning customer failures into reusable scoring rubrics, will compound credibility faster than vendors that only offer internal tooling.